Temporality in AI Planning and Natural Language Semantics

Mark Steedman: Informatics, Edinburgh

The paper discusses some resources from logic and artificial intelligence research that can be used to build knowledge representations that support common sense reasoning about events and times, so as to support a semantics for verbs, temporal adverbials and other natural language categories like tense, mood, and aspect. While these categories are commonly talked of as "temporal", the paper argues that they are primarily causal and teleological, and can be captured in a dynamic logic-based variant of the McCarthy/Hayes/Kowalski situation or event calculi called the Dynamic Event Calculus, using a fibered or bunched-implication logic including both intuitionistic and linear implication. When properly drawn up in the way that the natural language problem demands, such systems provide natural solutions to both the temporal knowledge representation problem and the semantics of temporality in natural language.

If I have time I'll discusses various criticisms of such calculi that have been mounted in the AI literature, based on supposed inability to handle various versions of the AI Frame Problem, including various forms of Ramification and Qualification Problems.